I am looking at moving forward with the Clan Scoreboard. As discussed with Leehar, due to the challenges involved in getting all the data input, especially with CL4, and also due to the time-window nature of clan scoring I think it makes sense to just move forward with the data we have.
So, I have been reviewing the F400 for implementation...
Once I parsed through all the graphics, adjustments, and tools to display and experiment, the basic formula seemed very simple. I want to confirm that this is the essence of it:
- Winning/Losing a Clan War wins/loses a Clan between 200 and 600 points depending on the Game Win % (200 points for 50.1% up to 600 points for 100%) - A weight is applied to the score for each war based on its recency with a time window - Weighted scores are averaged across time window
The surprising thing to me was that the scoring did not appear to depend on the prior record of the clans. So beating a weak clan scores as much as a strong Clan - with the exception of an adjustment for total noob clans.
I am looking at moving forward with the Clan Scoreboard. As discussed with Leehar, due to the challenges involved in getting all the data input, especially with CL4, and also due to the time-window nature of clan scoring I think it makes sense to just move forward with the data we have.
So, I have been reviewing the F400 for implementation...
Once I parsed through all the graphics, adjustments, and tools to display and experiment, the basic formula seemed very simple. I want to confirm that this is the essence of it:
- Winning/Losing a Clan War wins/loses a Clan between 200 and 600 points depending on the Game Win % (200 points for 50.1% up to 600 points for 100%) - A weight is applied to the score for each war based on its recency with a time window - Weighted scores are averaged across time window
The basics seem to be right, but Icepack will confirm
bigWham wrote:The surprising thing to me was that the scoring did not appear to depend on the prior record of the clans. So beating a weak clan scores as much as a strong Clan - with the exception of an adjustment for total noob clans.
Is this correct?
Nope, as you can see from this more detailed breakdown of scoring from 15 Jan:
Click image to enlarge.
Opposition score definitely impacts on the score that the individual clan gains from a war.
I am looking at moving forward with the Clan Scoreboard. As discussed with Leehar, due to the challenges involved in getting all the data input, especially with CL4, and also due to the time-window nature of clan scoring I think it makes sense to just move forward with the data we have.
So, I have been reviewing the F400 for implementation...
Once I parsed through all the graphics, adjustments, and tools to display and experiment, the basic formula seemed very simple. I want to confirm that this is the essence of it:
- Winning/Losing a Clan War wins/loses a Clan between 200 and 600 points depending on the Game Win % (200 points for 50.1% up to 600 points for 100%) - A weight is applied to the score for each war based on its recency with a time window - Weighted scores are averaged across time window
The basics seem to be right, but Icepack will confirm
bigWham wrote:The surprising thing to me was that the scoring did not appear to depend on the prior record of the clans. So beating a weak clan scores as much as a strong Clan - with the exception of an adjustment for total noob clans.
Is this correct?
Nope, as you can see from this more detailed breakdown of scoring from 15 Jan:
Click image to enlarge.
Opposition score definitely impacts on the score that the individual clan gains from a war.
honestly, it is not clear to me what is going on in that table. i see a lot of ratings, weights, %s everywhere, but there's too much going on to be sure of what all the calcs are made or what they mean.
my questions is, if TOFU beats KORT, winning 51% of games, is it any different to TOFU beating some other, lesser, clan by the same amount?
or, my more general question is... can someone cut through the clutter for me and give me the actual algorithm that is used to calculate clan scores?
bigWham wrote:honestly, it is not clear to me what is going on in that table. i see a lot of ratings, weights, %s everywhere, but there's too much going on to be sure of what all the calcs are made or what they mean.
my questions is, if TOFU beats KORT, winning 51% of games, is it any different to TOFU beating some other, lesser, clan by the same amount?
or, my more general question is... can someone cut through the clutter for me and give me the actual algorithm that is used to calculate clan scores?
All it does is give a break-down of what score is 'earned' for each war.
So tofu-kort was worth 1602 for Tofu winning(52%), & 1135 for Kort in losing (48%) (& which makes sense that kort didn't lose so many points since they lost narrowly to a strong clan)
So the win is worth more for tofu, then their winning 58% of the games against HH, where they only earned 1156 points (which makes sense, since HH is ranked so much lower)
& then finally all those individual scores against each opponent, are weighted using the factor of time mentioned earlier.
bigWham wrote:or, my more general question is... can someone cut through the clutter for me and give me the actual algorithm that is used to calculate clan scores?
bigWham wrote:or, my more general question is... can someone cut through the clutter for me and give me the actual algorithm that is used to calculate clan scores?
I honestly can't tell you what tweaks have been made, if any. We'll again have to wait on Ice
Seems like Ice is busy, but I think you received the info email from Ice, not sure how much it'll help with tabulation though
> > Greetings, > I'm sending / attaching the tool and data that is used currently to calculate clan rankings. > I'm happy to discuss on Skype ("uneek03") if you need some assistance. > > Attached are two items: > 1) Algorithm of 400 Performance Rating Calculator; and > 2) Notepad of all Clan Wars, Leagues, Results, and dates. > > It's a fairly simple system, right now the process is manually driven. > > Rating Calc Date: Enter the date you wish the calc to be done up to (for Nov 1st Rankings, you enter 11-01-13. > Compare to Date: Any date prior to 11-01-13 (or whatever date you entered). This adjusts the +/- arrows, but these aren't 100% accurate so they are done manually on our current spreadsheet. > Final Ranking - Min Total Weight Required to Include Clan: Current F400 is set @ 150 Weight Required > Accuracy Tabulation - Min Total Weight: Enter 100 min > Accuracy Tabulation - Min Games to include War: Enter 12 min > Accuracy Tabulation - Rating Lead Total Intervals: Enter 16 > Accuracy Tabluation - Intervals to tabulate: Enter 5 > Min Delta Per Win / Loss: Enter 200 > Min Delta Per Win Loss: Enter 600 > > Data Window (years): Enter 2 Years for F400 > > Graph (if desired) in years: Standard is 2 > Graph Clans Ranked: This calculator can rank 8 at a time. 1-8, 9-16, etc > > Copy and Paste the notepad records into the box, click calculate ratings. > > It will take a min and show the graph and results below. > > Currently, these informations are then entered into an excel and any display errors fixed and posted for clans viewing. > > Let me know if I can be of further assistance. > Thanks, > IcePack / Jason
bigWham wrote:or, my more general question is... can someone cut through the clutter for me and give me the actual algorithm that is used to calculate clan scores?
I honestly can't tell you what tweaks have been made, if any. We'll again have to wait on Ice
Seems like Ice is busy, but I think you received the info email from Ice, not sure how much it'll help with tabulation though
> > Greetings, > I'm sending / attaching the tool and data that is used currently to calculate clan rankings. > I'm happy to discuss on Skype ("uneek03") if you need some assistance. > > Attached are two items: > 1) Algorithm of 400 Performance Rating Calculator; and > 2) Notepad of all Clan Wars, Leagues, Results, and dates. > > It's a fairly simple system, right now the process is manually driven. > > Rating Calc Date: Enter the date you wish the calc to be done up to (for Nov 1st Rankings, you enter 11-01-13. > Compare to Date: Any date prior to 11-01-13 (or whatever date you entered). This adjusts the +/- arrows, but these aren't 100% accurate so they are done manually on our current spreadsheet. > Final Ranking - Min Total Weight Required to Include Clan: Current F400 is set @ 150 Weight Required > Accuracy Tabulation - Min Total Weight: Enter 100 min > Accuracy Tabulation - Min Games to include War: Enter 12 min > Accuracy Tabulation - Rating Lead Total Intervals: Enter 16 > Accuracy Tabluation - Intervals to tabulate: Enter 5 > Min Delta Per Win / Loss: Enter 200 > Min Delta Per Win Loss: Enter 600 > > Data Window (years): Enter 2 Years for F400 > > Graph (if desired) in years: Standard is 2 > Graph Clans Ranked: This calculator can rank 8 at a time. 1-8, 9-16, etc > > Copy and Paste the notepad records into the box, click calculate ratings. > > It will take a min and show the graph and results below. > > Currently, these informations are then entered into an excel and any display errors fixed and posted for clans viewing. > > Let me know if I can be of further assistance. > Thanks, > IcePack / Jason
Is there no way to feed off the calculator as is?
it is coded in javascript, and has to be converted into php. from what i can see, and i went and looked again, it does not appear to take prior record into account. maybe i am missing something glaringly obvious, and if so need to put on the right track. it is even theoretically possible that the code doesn't implement the intended algorithm. someone that knows the code or algorithm should be able to this pretty quickly.
bigWham wrote:or, my more general question is... can someone cut through the clutter for me and give me the actual algorithm that is used to calculate clan scores?
it is coded in javascript, and has to be converted into php. from what i can see, and i went and looked again, it does not appear to take prior record into account. maybe i am missing something glaringly obvious, and if so need to put on the right track. it is even theoretically possible that the code doesn't implement the intended algorithm. someone that knows the code or algorithm should be able to this pretty quickly.
alternately we can just make something new up.
From what I understand It's taken into account in the total of opponents ratings?
I don't think it'll be feasible to make something new up because it's accepted far & wide, we've seen previously that it can otherwise be a fairly contentious issue getting a ranking otherwise
bigWham wrote:or, my more general question is... can someone cut through the clutter for me and give me the actual algorithm that is used to calculate clan scores?
it is coded in javascript, and has to be converted into php. from what i can see, and i went and looked again, it does not appear to take prior record into account. maybe i am missing something glaringly obvious, and if so need to put on the right track. it is even theoretically possible that the code doesn't implement the intended algorithm. someone that knows the code or algorithm should be able to this pretty quickly.
alternately we can just make something new up.
From what I understand It's taken into account in the total of opponents ratings?
I don't think it'll be feasible to make something new up because it's accepted far & wide, we've seen previously that it can otherwise be a fairly contentious issue getting a ranking otherwise
i am probably missing something. there is a lot of code i was sent, but most of it does not pertain to the actual calculations. here is what i think is the central calculation:
var Delta = Math.round(MinWinGain + 2*(Math.max(Clan1WinFraction,1-Clan1WinFraction) - .5) * (MaxWinGain-MinWinGain));
Delta is the points given for the war Clan1WinFraction = % of games won in war by Clan 1 MinWinGain = 200 (default) MaxWinGain = 600 (default)
then these delta are summed over all wars and a weighting applied for how long ago the war was.
Couldn't see anything that took into account the record of your opponent in other wars (in other words some type of measure of how good your opponent was).
Is no one who understands the calculation still around?
bigWham wrote:or, my more general question is... can someone cut through the clutter for me and give me the actual algorithm that is used to calculate clan scores?
it is coded in javascript, and has to be converted into php. from what i can see, and i went and looked again, it does not appear to take prior record into account. maybe i am missing something glaringly obvious, and if so need to put on the right track. it is even theoretically possible that the code doesn't implement the intended algorithm. someone that knows the code or algorithm should be able to this pretty quickly.
alternately we can just make something new up.
From what I understand It's taken into account in the total of opponents ratings?
I don't think it'll be feasible to make something new up because it's accepted far & wide, we've seen previously that it can otherwise be a fairly contentious issue getting a ranking otherwise
i am probably missing something. there is a lot of code i was sent, but most of it does not pertain to the actual calculations. here is what i think is the central calculation:
var Delta = Math.round(MinWinGain + 2*(Math.max(Clan1WinFraction,1-Clan1WinFraction) - .5) * (MaxWinGain-MinWinGain));
Delta is the points given for the war Clan1WinFraction = % of games won in war by Clan 1 MinWinGain = 200 (default) MaxWinGain = 600 (default)
then these delta are summed over all wars and a weighting applied for how long ago the war was.
Couldn't see anything that took into account the record of your opponent in other wars (in other words some type of measure of how good your opponent was).
Is no one who understands the calculation still around?
I know you don't have much time, but I would suggest flipping through that thread if possible. It has a lot of bits that were discussed as the ranking was being developed, and it can explain the technicalities better than I can second-hand.
One thing I can say however, is that it might not hurt to see what output you can get from utilising that formula. It would help us to be able to identify what we may need to tweak in order to get a result that accurately reflect the F400 where it is currently widely accepted.
Edit: Ok, I've gained access to the console.
Whats happening with : var Clan1Score = new Number(Clan1RatingObject.Rating); var Clan2Score = new Number(Clan2RatingObject.Rating);
and how are they used?
From my understanding, for each clan war, the winner & a loser both get a score out of it. And I'm pretty sure it's not purely based on the 600-200 thing although it's a major factor. Like you though I can't isolate where the Clan 2 score is incorporated.
I'm thinking it may incorporate ( RatingRecords[RatingRecords.length] = new RatingRecord(Clan1Name,Clan2Score,Games,RecordDate,Clan2Name,Outcome1,WinLossOutcome1,Clan2RatingObject.Rating); RatingRecords[RatingRecords.length] = new RatingRecord(Clan2Name,Clan1Score,Games,RecordDate,Clan1Name,Outcome2,WinLossOutcome2,Clan1RatingObject.Rating);) ?
i am probably missing something. there is a lot of code i was sent, but most of it does not pertain to the actual calculations. here is what i think is the central calculation:
var Delta = Math.round(MinWinGain + 2*(Math.max(Clan1WinFraction,1-Clan1WinFraction) - .5) * (MaxWinGain-MinWinGain));
Delta is the points given for the war Clan1WinFraction = % of games won in war by Clan 1 MinWinGain = 200 (default) MaxWinGain = 600 (default)
then these delta are summed over all wars and a weighting applied for how long ago the war was.
Couldn't see anything that took into account the record of your opponent in other wars (in other words some type of measure of how good your opponent was).
Is no one who understands the calculation still around?
I know you don't have much time, but I would suggest flipping through that thread if possible. It has a lot of bits that were discussed as the ranking was being developed, and it can explain the technicalities better than I can second-hand.
One thing I can say however, is that it might not hurt to see what output you can get from utilising that formula. It would help us to be able to identify what we may need to tweak in order to get a result that accurately reflect the F400 where it is currently widely accepted.
Edit: Ok, I've gained access to the console.
Whats happening with : var Clan1Score = new Number(Clan1RatingObject.Rating); var Clan2Score = new Number(Clan2RatingObject.Rating);
and how are they used?
From my understanding, for each clan war, the winner & a loser both get a score out of it. And I'm pretty sure it's not purely based on the 600-200 thing although it's a major factor. Like you though I can't isolate where the Clan 2 score is incorporated.
I'm thinking it may incorporate ( RatingRecords[RatingRecords.length] = new RatingRecord(Clan1Name,Clan2Score,Games,RecordDate,Clan2Name,Outcome1,WinLossOutcome1,Clan2RatingObject.Rating); RatingRecords[RatingRecords.length] = new RatingRecord(Clan2Name,Clan1Score,Games,RecordDate,Clan1Name,Outcome2,WinLossOutcome2,Clan1RatingObject.Rating);) ?
i am probably missing something. there is a lot of code i was sent, but most of it does not pertain to the actual calculations. here is what i think is the central calculation:
var Delta = Math.round(MinWinGain + 2*(Math.max(Clan1WinFraction,1-Clan1WinFraction) - .5) * (MaxWinGain-MinWinGain));
Delta is the points given for the war Clan1WinFraction = % of games won in war by Clan 1 MinWinGain = 200 (default) MaxWinGain = 600 (default)
then these delta are summed over all wars and a weighting applied for how long ago the war was.
Couldn't see anything that took into account the record of your opponent in other wars (in other words some type of measure of how good your opponent was).
Is no one who understands the calculation still around?
I know you don't have much time, but I would suggest flipping through that thread if possible. It has a lot of bits that were discussed as the ranking was being developed, and it can explain the technicalities better than I can second-hand.
One thing I can say however, is that it might not hurt to see what output you can get from utilising that formula. It would help us to be able to identify what we may need to tweak in order to get a result that accurately reflect the F400 where it is currently widely accepted.
Edit: Ok, I've gained access to the console.
Whats happening with : var Clan1Score = new Number(Clan1RatingObject.Rating); var Clan2Score = new Number(Clan2RatingObject.Rating);
and how are they used?
From my understanding, for each clan war, the winner & a loser both get a score out of it. And I'm pretty sure it's not purely based on the 600-200 thing although it's a major factor. Like you though I can't isolate where the Clan 2 score is incorporated.
I'm thinking it may incorporate ( RatingRecords[RatingRecords.length] = new RatingRecord(Clan1Name,Clan2Score,Games,RecordDate,Clan2Name,Outcome1,WinLossOutcome1,Clan2RatingObject.Rating); RatingRecords[RatingRecords.length] = new RatingRecord(Clan2Name,Clan1Score,Games,RecordDate,Clan1Name,Outcome2,WinLossOutcome2,Clan1RatingObject.Rating);) ?
ok, i will revisit it when i get a chance.
i think i have worked this out. i implemented something that doesn't give me identical numbers, but similar. not sure where the differences lie as yet... one thing is i'm not sure if the data file i received with the javascript matches the database. anyway, should have something to share in the near future.
bigWham wrote:i think i have worked this out. i implemented something that doesn't give me identical numbers, but similar. not sure where the differences lie as yet... one thing is i'm not sure if the data file i received with the javascript matches the database. anyway, should have something to share in the near future.
Yeah, those files are pretty old.
The differences are most likely due to the extent at which we've included clan league matches.
bigWham wrote:i think i have worked this out. i implemented something that doesn't give me identical numbers, but similar. not sure where the differences lie as yet... one thing is i'm not sure if the data file i received with the javascript matches the database. anyway, should have something to share in the near future.
Yeah, those files are pretty old.
The differences are most likely due to the extent at which we've included clan league matches.
bigWham wrote:i think i have worked this out. i implemented something that doesn't give me identical numbers, but similar. not sure where the differences lie as yet... one thing is i'm not sure if the data file i received with the javascript matches the database. anyway, should have something to share in the near future.
Yeah, those files are pretty old.
The differences are most likely due to the extent at which we've included clan league matches.
bigWham wrote:i think i have worked this out. i implemented something that doesn't give me identical numbers, but similar. not sure where the differences lie as yet... one thing is i'm not sure if the data file i received with the javascript matches the database. anyway, should have something to share in the near future.
Yeah, those files are pretty old.
The differences are most likely due to the extent at which we've included clan league matches.
Is it feeding of our database on the live site, or the older one from beta? Could be a possible reason why it may be missing some pieces of data?
It is from beta, so yes it is out of date. i can deploy something live soon, or perhaps port data over so we can see current data. does it make sense though, seem correct?
bigWham wrote:i think i have worked this out. i implemented something that doesn't give me identical numbers, but similar. not sure where the differences lie as yet... one thing is i'm not sure if the data file i received with the javascript matches the database. anyway, should have something to share in the near future.
Yeah, those files are pretty old.
The differences are most likely due to the extent at which we've included clan league matches.
Is it feeding of our database on the live site, or the older one from beta? Could be a possible reason why it may be missing some pieces of data?
It is from beta, so yes it is out of date. i can deploy something live soon, or perhaps port data over so we can see current data. does it make sense though, seem correct?
Yup, I don't have any major issues with it. Obviously there's the crucial element with regards to data entry which is messing with the results, but as a base it does seem to be very much on the right track.
Perhaps you'd want to implement code to say: where Clan Status = Hold, they are not displayed on the ranking? eg. where the likes of Agents of Chaos & Empire are inactive clans so strictly shouldn't be displayed.
The ranking itself is pretty close to what we have for the Jan F400 ranking top 10
(With the likes of Tsm & Atl not making the list while we have the aforementioned inactives) so it really does seem to be more of a question of the inputs being responsible for any differences, since the process (calculations etc) seem to be working fairly well for the stable outputs.
With regards to the output though, I think the graph, and breakdown of rankings month-by-month is surplus to immediate requirements specially if it contributes to the slow loading of the page? Better to just have the last column included on a landing page? And maybe just update it periodically (daily/weekly etc)
Just for the purpose of testing BW's calculations can I make the following suggestion. If it's relatively easy to do then why not strip out the league results from the test data. You could then compare the output against the F41, if everything matches then we can proceed with confidence knowing that the calcs are correct.
Just for the purpose of testing BW's calculations can I make the following suggestion. If it's relatively easy to do then why not strip out the league results from the test data. You could then compare the output against the F41, if everything matches then we can proceed with confidence knowing that the calcs are correct.
1) Your ranking is still listing EMP, and AoC as active clans. They have since been closed and ACE is the new clan. Those results should still count for purposes of the clans that they faced, but need to be removed from the display / ranking. 2) It appears the rankings compared do not give the same results. I looked at Feb 2014 for example: Scoreboard: Rank Clan Score ----- F41 Rank Clan Score 1. TOFU 1499 1. TOFU 1588 2. ACE 1409 2. ACE 1477 3. KORT 1257 3. KORT 1356 4. IA 1231 4. PACK 1330 5. AOC 1216 < Inactive 5. IA 1323 6. ATL 1190 6. ATL 1288 7. PACK 1190 7. OSA 1261 8. TSM 1167 8. TSM 1239 9. EMP 1158 < Inactive 9. MD 1207 10. AFOS 1168 10. LOW 1192
Similar order, though there were a few differences (plus the inactives). Once that is cleaned up, maybe we can get a new test top 10 and compare. But some had differences, what is the minimum weight criteria you are using to display the war only results? Or do you have no min weight restrictions? (that could explain some of it)
icePack
fac vitam incredibilem memento vivere Knowledge Weighs Nothing, Carry All You Can
1) Your ranking is still listing EMP, and AoC as active clans. They have since been closed and ACE is the new clan. Those results should still count for purposes of the clans that they faced, but need to be removed from the display / ranking. 2) It appears the rankings compared do not give the same results. I looked at Feb 2014 for example: Scoreboard: Rank Clan Score ----- F41 Rank Clan Score 1. TOFU 1499 1. TOFU 1588 2. ACE 1409 2. ACE 1477 3. KORT 1257 3. KORT 1356 4. IA 1231 4. PACK 1330 5. AOC 1216 < Inactive 5. IA 1323 6. ATL 1190 6. ATL 1288 7. PACK 1190 7. OSA 1261 8. TSM 1167 8. TSM 1239 9. EMP 1158 < Inactive 9. MD 1207 10. AFOS 1168 10. LOW 1192
Similar order, though there were a few differences (plus the inactives). Once that is cleaned up, maybe we can get a new test top 10 and compare. But some had differences, what is the minimum weight criteria you are using to display the war only results? Or do you have no min weight restrictions? (that could explain some of it)
icePack
The minimum weight variable, by which I presume you mean the one that is labeled "Final Ranking - Min Total Weight Required to Include Clan" in the interface, seems to only affect whether or not a clan is displayed in the output. There are also a whole bunch of parameters for estimating accuracy, but again these do not impact the score calculations themselves.
I am not sure what data you are using for your input or to what extent it precisely matches the data in the database. Does the data for every war, Date (for which i am using End Date) and war score match exactly?
Yes their results still need to be included but they still need to be removed from display.
I'm using the F400 war results list minus everything that is below 41 games (ie leagues). While if should be pretty close to yours, there will be some dates that differ by a day or so but that's not enough to show such drastic changes.
And yes the weight is display only but that's why I'm asking, if mines set to display the same. If your not using a min weight to display then I. Need to set mine to zero.
fac vitam incredibilem memento vivere Knowledge Weighs Nothing, Carry All You Can
IcePack wrote:Yes their results still need to be included but they still need to be removed from display.
I'm using the F400 war results list minus everything that is below 41 games (ie leagues). While if should be pretty close to yours, there will be some dates that differ by a day or so but that's not enough to show such drastic changes.
And yes the weight is display only but that's why I'm asking, if mines set to display the same. If your not using a min weight to display then I. Need to set mine to zero.
I am not using any weight to display. i also was not explicitly holding out wars with less than 41 games, though i was holding out leagues. i don;t know if there is a difference... however when i re-ran with wars < 41 held out the numbers are still different.
concerned this could take forever, particularly when we are working from two different datasets... even they "should" be the same, chances are they are not. the F400 list you speak of... is that the one in the dropbox? if so, i see considerable differences on first look... for example that file seems to only have 1 war ending in 12/2013, whereas the database has quite a few.
perhaps i can run a test by exporting the database wars to a file in your format.
